Research map / 3D vision / Human pose estimation
Scaling 3D Compositional Models for Robust Classification and Pose Estimation
Xiaoding Yuan, Guofeng Zhang, Prakhar Kaushik, Artur Jesslen, Adam Kortylewski, Alan Yuille
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| sample N=32 vertex features per category for cross-category loss instead of full contrast (ablation) | NOVUM · ImageNet3D+ efficiency | GFLOPS: 61.2 -> 2.71 (-96%) | Improved |
| sample N=32 vertex features per category for cross-category loss instead of full contrast (ablation) | NOVUM · ImageNet3D+ efficiency | training time: 66.8h -> 9.3h | Improved |
| sample N=32 vertex features per category for cross-category loss instead of full contrast (ablation) | NOVUM · ImageNet3D+ efficiency | accuracy: 85.7 -> 86.5 | Improved |
| reduce sampled features from N=64 to N=32 (ablation) | 3D-CompNets · ImageNet3D+ efficiency | accuracy: 86.3 -> 86.5 | Improved |
| add grouped neural vertex and dynamically weighted contrastive learning to NOVUM | NOVUM ResNet50 · ImageNet3D+ classification synthetic | accuracy: 68.8 -> 69.3 | Improved |
| use 3D-CompNets instead of DINOv2 head | DINOv2 ViT-B/14 · ImageNet3D+ classification synthetic | accuracy: 72.3 -> 74.9 | Improved |
| use 3D-CompNets instead of DINOv2 head | DINOv2 ViT-B/14 · ImageNet3D+ classification corrupted | accuracy: 52.4 -> 59.2 | Improved |
| use 3D-CompNets instead of DINOv2 head | DINOv2 ViT-B/14 · ImageNet3D+ classification occluded | accuracy: 41.7 -> 43.7 | Improved |
| use 3D-CompNets instead of DINOv2 head | DINOv2 ViT-B/14 · Unseen categories pose estimation | accuracy: 38.9 -> 45.5 | Improved |
| use 3D-CompNets instead of DINOv2 head | DINOv2 ViT-B/14 · ImageNet3D+ pose estimation IID | accuracy: 56.9 -> 57.6 | Improved |
| use 3D-CompNets instead of NOVUM in pose estimation under corruption | NOVUM · ImageNet3D+ pose estimation corrupted | accuracy: 44.5 -> 45.7 | Improved |
| switch backbone from ResNet50 to ViT-B/14 for 3D-CompNets | 3D-CompNets · ImageNet3D+ classification IID | accuracy: 86.5 -> 93.5 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- NOVUM — NOVUM: Neural Object Volumes for Robust Object Classification (ECCV 2024) · better on 7 of 7 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- NOVUM: Neural Object Volumes for Robust Object Classification (ECCV 2024 · similarity 0.89)
- Neural View Synthesis and Matching for Semi-Supervised Few-Shot Learning of 3D Pose (NeurIPS 2021 · similarity 0.87)
- NeMo: Neural Mesh Models of Contrastive Features for Robust 3D Pose Estimation (ICLR 2021 · similarity 0.87)
- Common3D: Self-Supervised Learning of 3D Morphable Models for Common Objects in Neural Feature Space (CVPR 2025 · similarity 0.85)
- 3DCoMPaT200: Language Grounded Large-Scale 3D Vision Dataset for Compositional Recognition (NeurIPS 2024 · similarity 0.85)
- Source-Free and Image-Only Unsupervised Domain Adaptation for Category Level Object Pose Estimation (ICLR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
